raidr_cli

Reconstruct a working web app from an raidr capture bundle — plus the agent skill that drives the reconstruction.

bun add -g @sudobility/raidr_cli

Requires Bun ≥ 1.2. The CLI ships as TypeScript and runs under Bun directly.

Usage

raidr reconstruct <bundle.zip|dir> --out <dir> [--replay]
raidr token <apiHost> [--print]
raidr install [--claude] [--codex] [--agents] [--all]
raidr uninstall [--claude] [--codex] [--agents]

raidr install symlinks the reconstruct skill into the personal skills directory of a coding agent — .claude/skills, .codex/skills, or the shared .agents/skills path. The skill itself is runtime-agnostic: it drives a shell binary and reads JSON, so it works the same in Claude Code, Codex, Gemini CLI, and Copilot CLI.

raidr token <apiHost> gets a site's auth token for raidr's hosted MCP server without copying it out of DevTools. It opens the site's sign-in page in a real browser window on your machine (installed Chrome, then Edge, then Playwright Chromium) with a persistent profile in ~/.raidr/browser, so you sign in once. When a request carrying your token to a signed-in-only endpoint succeeds, the window closes and the token is saved to ~/.raidr/config.json under siteTokens (mode 0600). Your password never leaves the browser. --print also writes the token to stdout:

SITE_TOKEN=$(bunx --package @sudobility/raidr_cli raidr token api.example.com --print)

How the site sends its token comes from the host's API doc on raidr.app, fetched with the apiKey in ~/.raidr/config.json; without one, pass --login <url> --style bearer|header|cookie (plus --header-name, --cookie-name, --user-path). Exit code 2 means the window was closed before you were signed in.

The split: CLI does the deterministic work, the agent does the judgment

raidr reconstruct performs every stage that has a right answer — source-map recovery, bundle unpacking, JSON Schema inference, API and route modelling, project scaffold, and a Hono replay server that serves the captured traffic back. Each stage writes a JSON artifact to disk.

The reconstruct skill reads those artifacts and does only the work that needs judgment: implementing components, naming things, wiring routes. Its governing rule is that the bundle is evidence, not inspiration — everything written must trace to something captured, and where the bundle is silent it says so.

That boundary is what keeps the system testable. Everything the CLI does is covered by bun test; only genuinely model-shaped work lives in prose.

Why this is a separate repository

raidr_processor is imported by raidr_extension, which ships into a Chrome MV3 bundle. Its defining constraint is that it performs no I/O — no filesystem, no DOM in its tsconfig lib — which is what keeps it trivially testable and safe to bundle for the browser.

A CLI is the opposite: it needs fs, path, process, and zip extraction. Putting that in raidr_processor would place Node-only code in the dependency graph of a browser artifact and would end the mechanical enforcement of that purity.

So the dependency shape is a diamond, not a chain:

raidr_processor              pure: bundle format, redaction, coverage, inference
   ├── raidr_extension   browser: CDP capture, offscreen buffer, side panel
   └── raidr_cli         node: unzip, filesystem, codegen + the reconstruct skill

The two consumers never see each other.

Development

bun install
bun run typecheck
bun test
bun run fixtures:build      # build the sample apps
bun run fixtures:capture    # capture them into fixtures/bundles/

The raidr project

| Repository | Role | |---|---| | raidr_processor | Bundle format and pure analysis | | raidr_extension | Chrome MV3 extension that performs the capture | | raidr_cli | Reconstruction CLI and the agent skill — this repo | | raidr_web | Landing site |

License

BUSL-1.1 — see LICENSE.md.